class: top, left, inverse, title-slide # Untangling biodiversity changes
across a continuum of spatial scales ##
PhD Presentation ###
PhD candidate: François Leroy
Supervisor: Petr Keil ### Czech University of Life Sciences
Prague --- # Biodiversity changes are scale dependent * **Global** biodiversity is **declining** -- * **Local, regional or national** trends are **not always** similar <img src="data:image/png;base64,#images/litrev.jpg" width="2604" height="430px" /> --- # Biodiversity changes are scale dependent * **Global** biodiversity is **declining** * **Local, regional or national** trends are **not always** similar <br><br><br><br> .center[ ### `\(\Rightarrow\)` Dynamic processes (*i.e.* colonization, extinction, turnover...) vary with spatial scales ] --- class: inverse, center, middle ## Biodiversity trends have to be assessed across spatial (and temporal) scales --- # Scales * **Spatial grain** <br><br><br> .center[ <img src="data:image/png;base64,#images/spatialgrain.jpg" width="1831" /> ] --- # Scales * **Temporal grain** <br><br><br> .center[ <img src="data:image/png;base64,#images/temporalgrain.jpg" width="80%" /> ] --- # Biodiversity data * One dataset express the biodiversity at its specific spatial and temporal grains * Data heterogeneity in spatial and temporal grains and extent * Lack of data -- <br><br> ## Problem: With the actual data, it is not straight forward to assess biodiversity trends for a continuum of spatial scales --- # Spatial aggregation **Jarzyna *et al* (2015)** <img align="right" src="data:image/png;base64,#images/jarzyna2015.jpg" width="75%"/> <br><br><br><br><br><br><br> **Chase *et al.* (2019)** .center[ <img src="data:image/png;base64,#images/chase_2019.PNG" width="40%" /> ] --- # Model .pull-left[ * Use biodiversity data with heterogeneous: **spatial grain, temporal grain, location, spatial extent and temporal extent**. <br><br> * Use this component as **covariates** to predict species richness at desired (spatial & temporal) grain and location (in space & time) ] .pull-right[ <img src="data:image/png;base64,#images/keil_chase_title.PNG" width="1529" /> .center[ <img src="data:image/png;base64,#images/keil_chase_2022.PNG" width="75%" /> ] ] --- # Model <br><br> **In practice:** ``` treeBasedModel(species richness ~ area, -> Species-area relationship temporal grain, -> Species-time relationship latitude, -> Location in space longitude, -> Location in space date) -> Location in time ``` <br> Tree based models: the flexibility grasps the interactions between **species area/time relationship** and their location in space and time. The species-area and species-time relationships allows to down/upscale species richness -- <br> .center[ ### `\(\Rightarrow\)` We need data at different spatial and temporal grains ] --- # Atlas dataset .pull-left[ .center[**Temporal scales**] 3 time periods, 3 different time spans: * M2 = 1985-1989 (**5 years**) * M3 = 2001-2003 (**3 years**) * M4 = 2014-2017 (**4 years**) ] .pull-right[ .center[**Spatial scales**] Large scale dataset. Ranging from less than **100 Km** `\(^2\)` to **80 000 Km** `\(^2\)` (the entire Czech Republic) <img src="data:image/png;base64,#images/spatialgrain_atlas.jpg" width="1608" /> ] -- <br><br> .center[ ### `\(\Rightarrow\)` The model homogenize the temporal grain and the sampling effort ] --- # Breeding bird survey (BBS) dataset
--- # Breeding bird survey (BBS) dataset .pull-left[ **Spatial scales:** very local **Temporal scales:** from 0.5 year to 10+ years .center[ <br><br> ### `\(\Rightarrow\)` The model predict species richness for missing years ] ] .pull-right[
] --- # Performance .pull-left[ **Atlas model** `\(XGBoost\)` `\(R^2 = 0.77\)` `\(MAE = 9\)` <!-- ``` --> <!-- species richness ~ area, --> <!-- temporal grain, --> <!-- latitude, --> <!-- longitude, --> <!-- date, --> <!-- sampling effort, --> <!-- shape, --> <!-- elevation) --> <!-- ``` --> <img src="data:image/png;base64,#images/obsvspred_atlas.JPG" width="90%" /> ] .pull-right[ **BBS model** `\(Random Forest\)` `\(R^2 = 0.74\)` `\(MAE = 10\)` <img src="data:image/png;base64,#images/obsvspred_bbs.JPG" width="90%" /> ] --- # Richness change across scales * For each spatial scale: predictions of species richness from 1987 to 2017 * Assessment of the species richness change per year <br><br> <img src="data:image/png;base64,#images/spatialgrain.jpg" width="1831" /> --- # Richness changes across scales .center[ <img src="data:image/png;base64,#images/srtrend.jpg" width="85%" /> ] --- # Richness changes across scales .center[ <img src="data:image/png;base64,#images/srtrend_details.jpg" width="1369" /> ] --- # Colonization, extinction, persistence across scales * Species richness change is the sum of these 3 processes * Assessment across scales --- # Colonization, extinction, persistence across scales .center[ ## Extinction <img src="data:image/png;base64,#images/extinction_concept.JPG" width="75%" /> .credit[modified from Keil et al. (2017)] ] --- # Colonization, extinction, persistence across scales .center[ ## Colonization <img src="data:image/png;base64,#images/colonization_concept.JPG" width="68%" /> .credit[modified from Keil et al. (2017)] ] --- # Colonization, extinction, persistence across scales .center[ <img src="data:image/png;base64,#images/CEP.JPG" width="150%" /> ] --- # Beta-diversity across scales .pull-left[ **Similarity index:** `\(jaccard = \frac{pers}{pers+col+ext}\)` <br><br> <img src="data:image/png;base64,#images/jaccard.JPG" width="100%" /> ] -- .pull-right[ **Dissimilarity index** `\(betasim = \frac{min(ext,col)}{pers + min(ext,col)}\)` <br><br> <img src="data:image/png;base64,#images/betasim.JPG" width="1903" /> ] --- # Conclusion This pattern: .center[ <img src="data:image/png;base64,#images/srtrend-copy.jpg" width="75%" /> ] Can be explained by the spatial scaling of dynamic processes: * `\(\nearrow\)` **persistence** with increasing spatial grain * different `\(\searrow\)` slope of **extinction** and **colonization** with increasing spatial grain * `\(\searrow\)` temporal turnover with increasing spatial grain --- # Conclusion .center[ ### `\(\Rightarrow\)` As colonization, extinction and persistence are scale dependent, we observe a scale dependency of biodiversity change ] -- <br><br> .center[ ### `\(\Rightarrow\)` Assessing biodiversity trends at national scale doesn't inform much about local dynamic and vice-versa ] -- <br><br> .center[ ### `\(\Rightarrow\)` Using heterogeneous dataset allows to model biodiversity at location and time where data is missing ] --- class: inverse, center, middle # Thank you for your attention .footnote[ Email: leroy@fzp.czu.cz Twitter: @FrsLry ]